Application of NARX neural network model for discharge prediction through lateral orifices

Abstract

Prediction of side flow from a main channel through lateral orifices mainly for irrigation and environmental purposes is of great importance in hydraulic engineering. Besides analytical and experimental studies, artificial neural network has been widely used for the estimation of the discharge coefficient of flow through sharp crested circular and square orifices. In the present research, a nonlinear neural network technique, NARX, is employed for accurate prediction of the discharge coefficient and compared to other neural network techniques recently used for the same purpose. Extensive set of previously published experimental data for both square and circular orifices were used for modeling the discharge coefficient. Based on the root mean square error, RMSE, the mean absolute error, MAE, and the correlation coefficient, R, the NARX model predicted the discharge coefficient for side orifices (square or circular) under both small and large orifices assumptions with a better accuracy than the other neural network models. Also, the existing linear regression models that relates the discharge coefficient to the predominant parameters are checked for their accuracy against NARX. The results showed the superiority of the present model over all the techniques available in literature.